Application of ICAMM to Impact-Echo Testing - On Statistical Pattern Recognition in Independent Component Analysis Mixture Modelling - page 108

Signal

(a) Equipment configuration scheme (b) Photograph of equipment setup

Fig. 5.1 Equipment setup for impact-echo experiments a Equipment configuration scheme b
Photograph of equipment setup

longitudinal axis. This can also be observed for the sensors 3 and 6, and sensors 7
and 8.

5.3 ICAMM for Learning Material Defect Patterns

This section includes an ICA mixture model applied to the impact-echo signals.
There are no previous references to the use of ICAMM in NDT applications. This
approach is based on a new model for the joint probability density of features that
are extracted from the impact-echo signals as mixtures of ICA models. Hence, an
optimum classification method that is based on ICA mixture modelling (ICAMM)
is implemented. This study demonstrates that the proposed method is able to
distinguish different kinds of defects such as cracks, holes, and combinations of
cracks and holes. A low-cost operating NDT procedure is included, and its results
are presented for 3D finite element simulations and lab specimens. Note that
deriving detailed information from a simple impact is a difficult field of applica-
tion. Therefore, any results that show some capability of discerning different defect
characteristics have great potential interest in material diagnoses.
In Sect. 5.3 , we demonstrate that ICA can be used to separate information of
material defects in impact-echo testing [ 13 ]. In this section, ICAMM is approached
from a physical model that is based on dividing the wave path propagation into
two parts: impact to point flaws, and point flaws to sensors. It is assumed that the
set of point flaws builds defective areas with different geometries, such as cracks
(small parallelepipeds), holes (cylinders), and multiple defects (a combination of
cracks and holes). Depending on the kind of defective area, the spectrum measured
by the sensors changes, which allows the kind of defect condition of the material to
be discerned. This study demonstrates that the spectrum of different kinds of
defective materials can fit into different ICA models.